Physically-based dimensionless features for pluvial flood mapping with machine learning

Project overview

This is a companion submission for code and data used in the manuscript titled "Physically-based dimensionless features for pluvial flood mapping with machine learning." It includes the data used to develop the conclusions of the aforementioned work as well as a jupyter notebook to calculate the features, traing the logistic regression models, and produce most of the figures. It does not include the GRASS GIS code used to transform the raw data into the feature format we use here. We are happy to assist our readers in the development of these GIS operations by reaching out to us directly.

Code

- `dimensional_vs_dimensionless.ipynb`: Compute dimensionless features, trains logistic regression models for dimensional and nondimensional cases, produces figures and writes them.- `utils.py`: Utility functions used in the jupyter notebook.  

Data

Feather files are a binary columnar data format optimized for use with data frames, particularly in the Python and R programming languages. They are part of the Apache Arrow project and are designed to be fast and efficient for both reading and writing operations.

In this project, a file like huc12_020301030503_features_derivs_local.ftr contains tabular data related to hydrologic unit codes (HUCs). These files store various features and derived metrics for a specific hydrologic unit at a particular resolution (e.g., 10 meters). In this project, the resolution used is 1/3 arc-second which is approximately 10 meters, based on the Digital Elevation Models sourced from USGS 3DEP. The files are organized by HUC12 code and the scale used to calculate the features as explained in the corresponding publication. 

These feather files are read using the pandas library and contain tabular data with the following columns, which correspond to the definitions in Table 1 of the corresponding manuscript:- `x_pix` and `y_pix`: The integer pixel coordinates of the feature within the watershed raster grid. These are used to plot the results but are insufficient to place the results on a map.- `z_d`: The height above nearest drainage (HAND) value at the pixel location in meters.- `y_d`: The distance to the nearest drainage in meters.- `So`: The channel bed slope in meters per meter.- `no`: The Manning's roughness coefficient for the channel bed.- `ac`: The contributing area in square meters.- `A`: The channel cross-section in square meters.- `rh`: The hydraulic radius in meters.- `Ks`: The saturated hydraulic conductivity in meters per second.- `K`: The conveyance in cubic meters per second.- `Kvel`: The conveyance velocity in meters per second.- `Q_runoff_iuh_{interval}`: The runoff volume in cubic meters per second for the specified interval.- `heatmap_pluvial`: The exceedance probability of pluvial flooding. These are derived from two-dimensional HEC-RAS modeling of the watersheds of interest.